The Transfer of Resources From Middle-Aged Children to Functionally Limited Elderly Parents: Providing Time, Giving Money, Sharing Space

Transcription

1 Copyright 1999 by The Gerontological Society of America The Cerontologist ol. 39, No. 6, The transfer of resources from middleaged children to their functionally limited elderly parents is addressed from the perspective of the children who may allocate time to caregiving, share household space, and give money to parents. A simultaneousequations model estimates the extent to which the three modes of transfer are interdependent, given the parents' needs for resources and the children's ability to provide them. Caregiving is the primary mode of resource transfer and is of overriding importance for individuals who depend daily on help from other persons. Coresidence and financial assistance complement direct human assistance and, at the margin, have a substantially large effect on caregiving time. It is, therefore, relevant to consider all modes of transfer in order to better understand how families accommodate the needs of their frail and disabled members. Key Words: Caregiving, Coresidence, Financial assistance, Paid work The Transfer of Resources From MiddleAged Children to Functionally Limited Elderly Parents: Providing Time, Giving Money, Sharing Space Rachel F. Boaz, PhD, 1 Jason Hu, MBA, 2 and Yongjia Ye, PhD 2 The role of adult children in assisting functionally limited elderly parents has been receiving considerable attention because an increasing number of elderly parents survive to a very advanced age and, thus, are prone to frailty and functional limitations. The issue of filial responsibility has been addressed frequently from the perspective of the parents. This study addresses the transfer of resources to parents from the perspective of the children who may provide caregiving time, share living space, or give money. It investigates the interdependence among all three modes of transfer, given the parents' specific needs for resources and the children's ability to provide them. With increased longevity, the prevalence of threegeneration families is common. Nearly three out of Four persons in the age range of 5054 have at least one living parent and one living child (Soldo, 1996). At the time in their lives when called upon to assist frail and disabled parents, middleaged sons and daughters may still be paying for the education of their own children and otherwise assisting them during the early phases of adulthood. To support the succeeding generation, the middle generation can be expected to continue to work for pay. To help the preceding generation with daily activities of personal care and household tasks, members of the middle This study was supported by Grant #HS8232 from the Agency for Health Care Policy and Research. Linda A. Siegenthaler was the project officer and her encouragement is gratefully acknowledged. Address correspondence to Rachel F. Boaz, Graduate Center, Department of Economics, City University of New York, 365 Fifth Avenue, New York, NY rboaz@gc.cuny.edu department of Economics, City University of New York. generation have to supply caregiving services and, therefore, might cut down on market work; or, alternatively, they might continue to be engaged in market work and give money to their parents for paid help. Hence, the commitment to market work could affect not only the ability to transfer resources but also the mode of transfer. But, thus far, research on the role of labor market activity has been confined to the relationship between time devoted to caregiving and time spent in paid work. Several studies analyzed the effect of employment on caregiving. WhiteMeans and Chiang (1991) examined the effect of wages on hours of caregiving and found a negative effect at the time of the initial contact but not at the time of the 6month followup. Boaz and Muller (1992) determined that fulltime employment reduced caregiving to persons disabled in activities of daily living (ADLs) by 20 hours per week, but caregiving time had only a negligible effect on fulltime employment. Doty, Jackson, and Crown (1998) found that the reduced caregiving time of employed female primary caregivers is offset by secondary unpaid and paid helpers, which means that the elderly care recipients do not necessarily experience unmet need for help in the community. Other studies analyzed the effect of caregiving responsibilities on market work. Muurinen (1986) found that primary caregivers of terminally ill patients suffered a loss of income due to their withdrawal from the labor market or reduced hours of paid work. Stone and Short (1990) determined that caregivers of longterm disabled and frail elderly persons accommodated their market work to their caregiving responsibilities by rearranging 648 The Gerontologist Downloaded from

2 work schedules, taking time off without pay, or reducing hours of paid work. Ruhm (1996) determined that caregiving reduced employment by married but not by unmarried women. Wolf and Soldo (1994) found that caregiving by married women had a negative effect on their employment but the relevant coefficients are not statistically significant. Ettner (1996) determined that the effect of caregiving on paid work was conditioned on living arrangements; caregiving reduced daughters' employment when parents and daughters were living in different households but did not do so when they shared living quarters. Coresidence can facilitate caregiving by saving travel time for the caregiver. In general, both elderly parents and their children prefer living apart. According to a 1986 survey by Louis Harris and Associates (Harris and Associates, 1986), of all the elderly persons living alone, 86% stated their preference for this arrangement; and data from the 1990 Census show that only 15% of all elderly persons were living with adults other than a spouse (NIA, 1993). Pezzin, Kemper, and Reschovsky (1996) determined that subsidized paid help had enabled elderly individuals to continue living alone. But, in the absence of such help, functional limitations change this preference. BoerschSupan (1990) showed that the transition to sharing the household with a helper occurred because of limitations in personal care activities. BoerschSupan, Hajivassiliou, Kotlikoff, and Morris (1992) showed that elderly persons had a very strong preference for living with children to avoid residence in a nursing home. Soldo, Wolf, and Agree (1990) determined that coresidence of a widowed or divorced mother with her children depends on her functional status and on the number and gender of her children; mothers are more likely to reside with their children if they have only one child than if they have several children, and if this only child is a daughter rather than a son. Kotlikoff and Morris (1990), studying both elderly parents and their children, concluded that the children were less willing than their parents to share living quarters. Financial assistance to elderly parents, which could pay for human assistance, is a mode of resource transfer that received only scant attention because such transfers are rare. Intervivo financial support is more frequently provided to the succeeding generation on the threshold of adulthood. Summarizing the literature, Soldo and Hill (1993) stated that national estimates from the late 1980s indicate that only 5% of children gave money to parents or parentsinlaw over a 12 month period and that such assistance was provided to parents who were disabled at the end of their lives. Among sons and daughters aged 5161 in 1992, 9% transferred $500+ over a year's period (McGarry & Schoeni, 1995). Much of the literature on the transfer of resources between generations has focused on motivation, namely, whether transfers are a form of implicit exchange or can be attributed to altruism on the part of the donor. In the former case, donor and recipient "bargain" about whether the transfer should take place and at what level, and the donor of services or money anticipates a form of payment (e.g., an inheritance). In the latter case, the donor decides on making the transfer without expecting a return. Soldo and Hnl (1993) point out that some empirical findings support the hypothesis of exchange whereas others support the hypothesis of altruism, an inconclusiveness that might be due to the heterogeneity of donors, some of whom are motivated by altruism whereas others might expect a reward. A couple of more recent studies found that the needs of recipients motivate the transfer of resources. An analysis by McGarry and Schoeni (1995) supported the conclusion that transfers from living parents to their children and transfers from adult children to their elderly parents were based on the recipients' needs. Sloan, Picone, and Hoerger (1997) determined that the supply of time by children to disabled elderly parents did not appear to have been guided by the exchange motive of anticipating an inheritance. This study is concerned with the donors' ability to provide resources rather than with their motivation. Controlling for the donors' own economic resources and the other expected claimants on these resources, the study investigates the extent of interdependence among all three modes of resource transfers because interdependence implies that it is relevant to consider all modes of transfer simultaneously in order to better understand how families accommodate the needs of their frail and disabled members. Yet this interdependence has not yet received much attention. Further, because the transfers flow from the pooled resources of the donor household, this study treats transfers as a household activity, a departure from the customary approach to transfers as an individual activity. The next section discusses conceptual issues. It is followed by sections that specify the regression model and present empirical results, and it concludes with a discussion and an interpretation of the findings. Conceptual Issues The three modes of transfer (i.e., the provision of time, the giving of money, and the sharing of space) are not likely to be independent of each other. Their interdependence is explained by showing how a frail and disabled person can be helped. For a person who cannot perform basic functions of daily living without human assistance, caregiving services are of overriding importance. Hence, central to the transfer of resources is the provision of time (measured by hours of caregiving). The sharing of space facilitates caregiving by saving travel time and, therefore, complements caregiving. A person who shares living quarters with a disabled parent is usually a primary caregiver. Financial assistance to pay for needed human assistance can serve as a substitute for caregiving time that is not provided 'or as a supplement to the time that is provided. Labor market activity by the donor affects the ability to transfer resources. It can facilitate one mode of transfer and inhibit another. When sons or daughters do not work for pay, they can provide more caregiving time and less financial assistance than when they do. Conversely, the more time spent in paid work, the ol. 39, No. 6, Downloaded from

3 less time can be allocated to caregiving and the greater the ability to transfer money. Commitment to market work might also affect the decision on coresidence. A caregiver who is engaged in labor market activity might decide to share living quarters with a parent because coresidence is a timesaving device. Thus, the labor market activity of donors is an essential determinant of the ability to make transfers; it affects the mode of transfer ana the amounts transferred and, in turn, is affected by decisions on transfers. The interdependence among the three types of resources and their link to the labormarket activity of the donor call for a simultaneousequations model with four structural equations. Such a model is presented in Table 1; it satisfies the orderandrank conditions of identification. All four endogenous variables are righthandside (RHS) variables. The criterion for including exogenous variables in any structural equation is whether these variables can be expected to have a direct effect on the lefthandside (LHS) variable of that equation; the exclusion restrictions follow from the variables that are included in each equation, based on this criterion. The indirect effects are captured through the RHS endogenous variables. The exogenous variables (which are discussed in greater detail in the next two sections) are the characteristics of parents (recipients) and children (donors). The parents' characteristics indicate their needs for help and whether they have other family members (a spouse and other children) who could assist them. These characteristics have a direct effect on each of the three equations that estimate transfermodes; they affect "paid work" by sons and daughters (the donors) indirectly through the effects of the three RHS endogenous variables of the "paid work" equation. The ability of sons and daughters to transfer resources to parents is indicated by their commitment to market work, their own children, their wealth, and their health. As shown by Mincer (1974) and subsequent studies, the major determinants of paid work are schooling and age; they can serve as instrumental variables for compensation from work because these variables are highly correlated with labor market earnings. Fungible wealth (represented by income from assets) enhances the ability to provide financial assistance to parents but weakens the commitment to market work. The donor's own children might compete with their grandparents for resources. The type of resources that these children would need depends on their ages and, therefore, their ages affect the mode of resource transfer to their grandparents. The donor's general health status can affect both labor market and caregiving activities. Difficulty with mobility points to a specific aspect of health status that has a direct effect on helping parents with personal care activities such as dressing, bathing, and assisting a person with getting in and out of a bed or chair. Problems with mobility are less likely to affect market work because, as stipulated by the Americans With Disability Act, most workplaces have to accommodate these problems. (The undesirable effects of multicolinearity are avoided by including these two healthstatus indicators in different equations.) ariables Table 1. A SimultaneousEquations Model of Resources Transferred to Disabled and Frail Parents A. 1Endogenous ariables Time (Y,) Money (Y 2 ) Space (Y 3 ) Paid work (Y 4 ) B. texogenous ariables B.1 B.11 B.12 B.13 B.2 B.21 B.22 B.23 B.24 B.25 B.26 B.27 Parents' characteristics Need for help Their other children Their marital status Characteristics of sons or daughters Highest level of schooling Age Race General health status Difficulty with mobility Assetincome Ages of own children Type of Resources Transferred Time Money Space Y Y Y v v v v Pair! I CtlJ Work Note: v = variable that is included in equation; y = symbol for endogenous variable. The Specification of the Regression Model The data source for the empirical analysis is the first wave of the Health and Retirement Survey (HRS) which, in 1992, interviewed 12,652 persons (respondents) who were members of 7,702 households. The selection criterion for inclusion in the HRS is that at least one person in each household was born between 1931 and 1941 (aged 5161 in 1992). The spouse or domestic partner of the ageeligible person might be age eligible or could have been born before 1931 or after Of all respondents, 9,824 (78%) were age eligible and 1,138 (9%) were older. Their ages suggest that their parents were likely to be old: 40% of their parents were between ages 7079 and 50% were aged 80+. (Detailed information on the HRS is provided in Juster & Suzman, 1995). The criterion for inclusion in the present study is the functional status of the parents. Functional status and chronic health problems in old age are based on such exogenous factors as genetic endowment and events earlier in a person's life (Smith & Kingston, 1997). HRS considers a parent to be functionally limited if he or she needs help with routine personal activities such as dressing, eating, or bathing and/or "cannot be left alone for more than one hour" (e.g., question E43). This is the only question about the parent's need for assistance with activities of daily living (ADLs); it differs from the usually detailed questions asked when disabled individuals are interviewed (e.g., in such surveys as the NLTCS or AHEAD). 650 The Gerontologist Downloaded from

4 The unit of analysis is the household (not the individuals in the household) because the household is the economic unit whose members pool the resources that are transferred to parents. The HRS distinguishes between onerespondent households and tworespondent households; 370 households of the former type and 1,298 of the latter qualify for inclusion in the sample because these are the households that have at least one parent who needs help with personal care activities. The study is based on the sample of tworespondent households. Although similar in concept, the two types of households have to be analyzed separately; the analysis of the tworespondent households has to take into account the interaction between the labormarket activity of the two respondents whereas there is no such interaction in a onerespondent household. This study is based on the more complicated and larger subset, that is, on the tworespondent households. Each of these households happened to be paired by respondents of the opposite sex; the households were not selected on the basis of this attribute. The four endogenous variables are: (a) hours of caregiving time, (b) the probability of giving money to a parent, (c) the probability of coresiding with a parent, and (d) the commitment to market work. The HRS questionnaire words the question about hours of help in terms of the assistance provided during a 12 month period preceding the interview. It specifies the lower limit as 100 hours because persons who provide fewer than 100 hours in a year are considered to have provided only token assistance and no regular help with the daily activities of personal care; they are not considered to be caregivers of persons who need daily help and/or cannot be left alone. HRS sets the upper limit at 5,824 annual hours (approximately 16 hours per day) for each respondent or 11,648 annual hours for a tworespondent household. The HRS asks about amounts of financial assistance that exceed $500 per year. However, because the information on dollar amounts is incomplete, the regression analysis specifies the probability that money has been given to a parent rather than tne amounts transferred. The HRS provides information on whether a parent resides witn the respondent in the same household or has other living arrangements. The strength of the commitment to market work is measured by the amount of time spent in paid work. Labor supply (hours of paid work) is not a continuous (linear or loglinear) variable because reported hours of work are clustered. Juster and Stafford (1991) point out that hours of work reported in surveys usually cluster around 3540 weekly hours as a convenient indicator of fulltime paid work. In the HRS, a substantial cluster occurs at exactly 40 hours per week for 52 weeks, that is, at 2,080 annual hours (27% of all male respondents and 21% of all female respondents in the sample reported this number of hours). Hence, 2,080 is considered the demarcation for fulltime employment. A distinction is made among three types of commitment to market work: no commitment (no paid work), some commitment (parttime paid work of fewer than 2,080 annual hours), and a major commitment (fulltime paid work of 2,080+ annual hours). This classification follows the observations by labor economists (e.g., Blank,1989) that parttime employment is a distinct mode of employment and is on a different track than fulltime employment (analogous to the difference between adjunct and tenuretrack faculty). The qualitative difference between parttime and fulltime employment is also demonstrated by the difference in eligibility for employmentrelated benefits and opportunities for promotion. Put differently, parttime and fulltime employment are not on the same continuum. Multinomial logit (MNL; rather than ordered logit) is the appropriate functional form to represent the different types of employment because MNL estimates a different coefficient vector for each employment category. (Ordered logit does not have separate coefficient vectors because all truly ordered categories are on the same continuum.) The exogenous variables represent parents' need for human assistance and their ability to obtain this assistance from other members of their family. The characteristics of the respondents are indicators of their ability to provide resources to their parents; the relevant characteristics are their labormarket activity, their wealth and health, and the other expected claimants on their resources. A parent needs extensive human assistance when he or she requires the constant presence of a caregiver and cannot perform the daily activities of personal care without assistance. As a measure of the burden on donor households, this study distinguishes between households with at least one parent who needs such extensive help and households in which no parent needs extensive help. Another measure of the burden on households is the number of parents who need help, which is specified by differentiating between households in which only one parent needs help and households in which more than one parent requires human assistance (the maximum is four parents). Ability of the parents to rely on family members other than the respondent lessens this burden; it is indicated by the parent's marital status and the number of other sons and daughters (the respondent's siblings). A parent who is married and/or has other children is less dependent on caregiving by the respondent and less likely to reside in the respondent's household. The respondent's propensity to engage in market work depends on the level of education, age, and health status. These variables are highly correlated with the level of compensation (wages, salaries, and benefits), which is a major determinant of participation in labormarket activity. Therefore, these variables can serve as instruments for compensation from paid work and avoid the problem of RHS endogeneity of the direct measures of compensation. The decision on whether to commit to market work and the type of commitment is also affected by the household's income from assets (nonlabor income) and by the ages of respondents' children. The ages of respondents' own children determine the type of support that these children need. When almost all children are young adults (ages 18+), the most likely resource they would need is money, especially if they have not yet completed their education; and they may also need space ol. 39, No. 6, Downloaded from

5 if they reside in their parents' household. Hence, the grandchildren can be expected to compete with their grandparents for the respondents' financial support and household space. Age 25 was selected as the demarcation between need for parent support and no need (or lesser need) for such support. The number of children under age 26 is specified as an indicator of the other potential claimants on the resources of the respondents. As presented in Table 1, the regression model specifies four structural equations. A twostage procedure is used to estimate each single equation. (System methods such as fullinformationmaximumlikelinood turned out to be intractable for a fourequation model.) In the first stage, each of the four endogenous variables is regressed on all exogenous variables of the model; the rationale (as clearly explained in Gujarati, 1988, pp ) is that every exogenous variable is expected to affect each endogenous variable either directly or indirectly (through its direct effect on another endogenous variable). Four secondstage (structural) equations specify the predicted values of the endogenous variables as RHS variables. The functional forms for the firststage and secondstage equations are: (a) a tobit equation for hours of human assistance (time), (b) a binary (ordinary) logit equation for the probability of providing financial assistance, (c) a binary (ordinary) logit equation for the probability of coresidence (sharing space), and (d) an MNL equation for estimating the commitment to market work. The unit of analysis is the household because transfers flow from the household's pooled resources. This approach differs from the customary one that focuses on the individual as the unit of analysis even when the individual is married and data are available for the spouse. The commitment of one spouse to market work is not independent from the commitment to market work of the other spouse (Killingworth, 1983, pp. 3034). Because each household happened to be paired by respondents of the opposite sex, the labormarket interaction between its two respondents is presented as genderbased interaction. The MNL categories of the variable "commitment to market work" are specified to reflect this interdependence, as follows: (a) both husband and wife work fulltime; (b) only the husband works fulltime (his wife does not work or works parttime); (c) only the wife works fulltime (her husband does not work or works parttime); (d) both spouses do not work; and (e) a residual category of the possible remaining interactions (a category that is not interpreted). Because the HRS oversamples racial and ethnic minorities, the sample observations are weighted to adjust for the oversampling so that the empirical results are applicable to the population represented by the sample of this study. Empirical results Table 2 shows the frequency distribution of tworespondent households by types of resources transferred. Although every household had at least one parent who needed help with personal care activities and/or who could not be left alone, only 29% of these Table 2. Number of Households Transferring Resources by Type of Resources Transferred Type of Transfer Time, space, and money Time and space Time and money Space and money Time only Space only Money only No transfer of resources All households No. of Sampled Households ,298 Proportion of Households Unweighted Weighted a Weighted proportions represent the population whereas unweighted proportions represent the sampled households. households transferred any resources over a 12month period. The most preferred modes of resource transfer were either only caregiving (12%) or only money (10%). A combination of time and money was rare (3%). But even more rare was the sharing of space, either as the only mode (1.6%) or in combination with time (1.8%). And less than 1% of households provided a combination of all three modes of transfers. The provision of any assistance was most likely when the parent needed extensive help (as defined earlier) and when the parent had no spouse. Compared with the entire cohort of HRS respondents, transfers were more common among respondents with functionally limited parents. McGarry and Schoeni (1995) show that, for the larger denominator of all HRS respondents with living parents, 7% provided time and 9% provided money; the comparable proportions for the subset of households with at least one functionally limited parent were 17% and 14%, respectively. Table 3 presents descriptive statistics (weighted arithmetic means and proportions) of regression covariates that represent household characteristics and the personal characteristics of their male and female respondents. For each transfer mode, this table flags the statistically significant differences between households that transferred resources and households that did not. Households that provided time differed from the ones that did not in having proportionally more parents with extensive needs for help and proportionally fewer parents who were married. They also had proportionally more respondents who did not work for pay and proportionally fewer men who were employed fulltime. Households that provided financial assistance differed from the ones that did not in having higher levels of assetincome and proportionally more women who were engaged in market work fulltime. They also had a somewhat higher proportion of parents with extensive needs for human assistance and a smaller proportion of married parents. Although marriage per se does not confer a better financial position, data from the Current Population Survey show that, among the 652 The Gerontologist Downloaded from

6 Table 3. The Personal Characteristics of HRS Respondents With Functionally Limited Parents by Type of Resources Transferred to Parents (TwoRespondent Households) Provide time? Provide money? Provide space? Characteristics Yes No Difference Yes No Difference Yes No Difference Weighted means (adjusted for oversampling) A. 1. Respondents Paid Work: Men Fulltime Parttime No paid work Paid Work: Women Fulltime Parttime No paid work *.023.o * * *.142* 2. Schooling: Men Grade school Some high school High school completed Some college College completed Post college Schooling: Women Grade School Some high school High school completed Some college College completed Post college * * * * *.214,143, ,167, ,083, ** Perceived Health: Men Excellent ery good Good Fair or poor Perceived Health: Women Excellent ery good Good Fair or poor * ,369, ** ** Difficulty With Mobility Men Women ** 5. Age <54: Men Age <54: Women B. 1. Households Asset Income ($) a None 11,000 1,0015,000 5,00115,000 15, ** * * Children Under Age 26 No children whatsoever No children under children under children under ** 3. Only 1 Parent Needing Help At Least 1 Parent Needing Extensive Help ** (Table continues on next page) ol.39. No fim Downloaded from

7 Table 3. The Personal Characteristics of HRS Respondents With Functionally Limited Parents by Type of Resources Transferred to Parents (TwoRespondent Households) {continued) Provide time? Provide money? Provide space? Characteristics Yes No Difference Yes No Difference Yes No Difference Weighted means (adjusted for oversampling) 5. At Least 1 Parent Is Married 6. No. of Other Children That Parents Have 7. NonWhite No. of Sampled Observations Observations Adjusted for Oversampling of Minorities , , ** " *.095** , a This classification reflects the skewness of this distribution < p<.100; *.010 < p<.050; **p <.010. elderly population, married couples are better off than unmarried individuals (Social Security Administration, 1998). Sons and daughters were more likely to share living quarters with parents who were not married and/ or had few or no other children; these households had proportionally fewer employed respondents, presumably because these respondents were likely to have had major caregiving responsibilities. The regression analysis is designed to assess the interdependence among transfer modes and to determine the extent to which parents' needs and children's ability to be donors affect the transfer of resources. For the sample of 1,298 households, Table 4 presents the secondstage equations that estimate the relationship among the three modes of transfer, controlling for the effects of the relevant characteristics of the donors and the recipients. The three dependent variables corresponding to these three transfer modes are hours of caregiving, the probability of transferring money, and the probability of sharing space. A fourth equation, whose dependent variable is market work, is not shown here because market work is not a resource that is transferred; its role in this model is that of a RHS endogenous indicator of the ability to transfer resources. The standard errors of the coefficients in secondstage equations are biased downward; hence, the observed level of statistical significance is overstated. To take account of this problem, the analysis disregards the 10% significance level, regards the 5% level as only a weak statistical significance, and interprets the 1% level as statistically significant. [We know of no program that would adjust the standard errors of a secondstage equation that has three RHS predicted variables that are obtained from three different firststage equations. Such an adjustment requires a combination of these three equations and, therefore, is not doable. LIMDEP 7.0 (Econometrics Software, Inc., New York) has a program for a model that has one firststage equation.] The RHS endogenous variables provide estimates of the interdependence among the three modes of resource transfer and the effect of market work on specific transfer modes. Household characteristics include parents' attributes, the respondents' own children, and the household's assetincome. Each of the households happens to have a male respondent and a female respondent (as explained in the previous section); therefore, gender characteristics are considered household characteristics and are shown separately for male and female spouses (partners). The regression coefficients are interpreted as the effects of RHS variables on the transfer of resources from the donor (sons/ daughters) generation to the recipient (parents) generation when the older generation has at least one functionally limited member. The interdependence among transfer modes is unidirectional in that financial assistance and coresidence affect caregiving time but caregiving has no reciprocal effect on the transfer of money or on the sharing of living space. Households that provided financial assistance also provided 502 more annual hours of caregiving time than households that did not provide such assistance. The few households that shared living quarters with a parent provided close to 3,000 additional caregiving hours per year compared with the many more households that did not share space. When a functionally limited parent resides in a respondent household, at least one member of the household is likely to have a major responsibility for caregiving. Coresidence and financial assistance do not appear to be interdependent. The transfer of money to parents did not affect the probability of sharing space with them, and coresidence did not affect the provision of financial assistance. In theory, the commitment to market work can be expected to affect every mode of transfer. However, the analysis shows that market work affected only the amount of caregiving time and had no effect on the probability of financial assistance or the probability of coresidence. Specifically, compared witn no market work, fulltime employment by both husband and wife reduced caregiving by 344 annual hours. When only 654 The Gerontologist Downloaded from

8 Table 4. Resources Transferred to Parents: Time, Money, and Space (SecondStage Weighted Regression Equations) Annual Hours of Help Financial Assistance Coresidence ariables Tobit Coefficients Marginal Effects 3 Logit Coefficients Marginal Effects Logit Coefficients Marginal Effects A. Predicted ariables 1. Hours of help.001 b.001 b 2. Financial assistance 3,306* Coresidence 19,716** 2, Paid work M fulltime & F fulltime M fulltime & F parttime or no work F fulltime & M parttime or no work Other combinations 2,267* 1,627* 582 2, B. Exogenous ariables 1. Households having only one parent needing help Households with at least one parent needing extensive help 1,317** Households with at least one parent who is married ** * No. of other children that parents have * ** Household's asset income ($) 11,000 1,0015,000 5,00115,000 15, * ** i i i i 6. Respondents' own children under age 26 C No children under children under children under Respondents having difficulty with mobility Men Women Respondents selfreported health status: fair or poor Men Women NonWhites ** Intercept 2,938** ** * Sigma 2,912** Notes: Reference categories for nonbinary variables: no paid work by both men and women (A.4),no asset income (B.5), no children whatsoever (B.6). The underlying firststage equations are available upon request. a The marginal effects are specified by McDonald and Moffitt (1980, Eq. 5). b Less than.0005 c Almost all own children are young adults (ages 18+). Their claims on household resources are for financial support and space accommodation. They do not compete with their disabled grandparents for help (time) with routine activities of daily living. Or put differently, cell size for the relevant group of minor children who would need such help is very small, especially when interacted with the number of children in the household < p <.100 (interpreted as not statistically significant);*.010 < p <.050 (interpreted as weak statistical significance);**p <.010 (interpreted as statistically significant). ol. 39. No Downloaded from

9 the husband was employed fulltime, caregiving was reduced by 247 annual hours, but caregiving by the household was not reduced when only the wife was employed fulltime. Market work had no statistically significant effect on providing financial assistance to parents or coresiding with them. Parents' characteristics have an effect on the transfer of resources by their children. Whereas the number of functionally limited parents in a household had no statistically significant effect on any transfer mode, a parent's need for extensive help prompted the households of their sons or daughters to provide 200 additional caregiving hours per year (compared to a situation when a functionally limited parent did not need extensive help). A parent's marital status and the number of other children (respondent's siblings) had statistically significant effects on reducing the probability of coresidence and financial transfers. Having a spouse reduced the probability of household sharing by 3.2% and the probability of financial support by 7.1%; having an additional child (a sibling of the HRS respondent) reduced the probability of coresidence by 0.6% and the probability of financial transfers by 0.8%. The ability of middleaged sons and daughters to assist their functionally limited parents depends on the household's fungible wealth and on the other expected claimants to its resources. A measure of such wealth is income from assets; a substantial level of assetincome increased the probability of financial transfers to parents by 8.7% when compared with a household that had no assetincome. Foremost among the other claimants are the donor household's own children, almost all of whom were young adults. These young adults might need financial support, and they might still reside in their parents' home. Thus, the grandchildren and grandparents might compete for space and money but, in effect, they did not. Having grandchildren under age 26 did not reduce the probability of financial assistance to their grandparents or coresidence with them; there is no statistically significant difference in giving money or sharing space between households that had no children and households that had any number of children under age 26. (One puzzling result is the effect of being nonwhite on the probability of transferring money to parents; contrary to expectations, being nonwhite rather than White increased this probability by 8.0%.) Discussion Intervivo transfers within families usually flow from middleaged members to their children; transfers from children to parents are very rare and occur mainly at the end of the parents' lives. The analysis is based on households in which at least one member is within the age range of When a son or a daughter has a spouse, the household can have four living parents. Included in this study are all households in which the ageeligible member has a spouse/partner (i.e., tworespondent households) and at least one parent cannot independently perform the daily activities of personal care and/or cannot be left unattended. The family life cycle suggests that, in their fifties, members of the "sandwich" generation are likely to have parents of very advanced age who are prone to frailty and disability, in addition to having children and even grandchildren; in other words, members of the middle generation can expect to have many claimants on their resources. This study controls for the ability of the sandwich generation to transfer resources as it investigates the interdependence among the three mode of resource transfer. Most studies of disabled elderly parents and their adult children focus on caregiving. Stone and Kemper (1989) estimated that fewer than one in four (23.6%) sons and daughters were active caregivers in Himes (1994) determined that, at any point in time, only few daughters are caregivers; but, over their lives, the probability of being a caregiver is high because, as parents' survival rates increase, the risk that they will become dependent on their children increases. The present study shows that, over a duration of one year, relatively few children provided caregiving time. On a household basis, 17% of the tworespondent households provided caregiving time in However, caution has to be exercised in concluding that family caregiving has declined over time because of the difference in the specified activities for which help is needed. The 1984 data have information on IADL and ADL disabilities whereas the 1992 data have information only on ADL disabilities; it is much easier to assist with household tasks than to help with the daily activities of personal care and, therefore, proportionally more children might be reported as caregivers in 1984 than in The provision of caregiving time is the primary mode of resource transfer. When tne transfers of money and space are counted in addition to the transfer of time, 29% of households transferred resources to parents. Coresidence has been studied as a facilitator of caregiving, but money transfers have received only scant attention although financial assistance enables a recipientparent to obtain paid help. The findings from this study suggest a unidirectional rather than a multidirectional interdependence among the types of resources that are transferred. Coresidence and financial support are associated with a substantial increase in the amount of caregiving time and, at the margin, complement caregiving. Households that share living quarters with parents are likely to have a major responsibility of caring for them. Financial support for paid help in addition to the unpaid help that is provided by family members signals a need for extensive human assistance by care recipients (Liu, Manton, & Liv, 1985); it is positively associated with caregiving time. Further, transfer modes appear to be targeted to parents' needs. The provision of caregiving time is increased when a parent is dependent on others for performing the daily activities of personal care and cannot be left alone. Coresidence is more likely when a parent has few or no other family members, such as a spouse and/or other children, with whom to share living quarters. Money transfers are more likely when a parent is not married because married elderly parents are frequently better off financially than unmarried elderly parents. 656 The Gerontologist Downloaded from

10 The modes of transfer also depend on the types of resources available to the donor households. Households are more likely to provide financial support when they have a substantial amount of income from assets, suggesting a substantial level of fungible wealth. Further, the likelihood of giving money to parents is not reduced by the other potential claimants (e.g., own children) on the household's resources. Market work was found to affect only the provision of help time. Fulltime paid work by both spouses reduces the household's caregiving more than if only one spouse is employed fulltime; but fulltime employment by only the husband is more likely to reduce the number of hours of help than fulltime employment by his wife. Conversely, when men and women are not engaged in market work, they provide more caregiving time than when they are. Although it has been amply documented that women have the major responsibility for caregiving, the caregiving households in this study display a pattern of human assistance in which men share this responsibility to a substantial extent. In 55% of all caregiving households, both spouses were caregivers; men were sole caregivers in 7% of these households, and women were the only caregivers in 38% of these households. (The information on household caregiving was, by design, obtained from the female respondent in the tworespondent household.) The findings suggest that it is relevant to consider all modes of transfer in order to understand how families accommodate the needs of their frail and disabled elderly members. Whereas the provision of caregiving services is the primary mode of transfer, almost half of the donor households provide financial assistance. On the other hand, the general trend away from sharing living quarters with elderly parents, which can be traced as far back as the 1940s (Schoeni, 1998), is also reflected in the very low rate of coresidence with elderly parents who are functionally limited. This finding implies that single parents have to find other assisted living arrangements when they are no longer able to live alone. To judge from the level of cash income available to individuals of advanced age (Social Security Administration, 1998), the majority of persons in this group cannot afford to pay for assisted living in the community without financial support from their children or subsidies from publicsector funds. Thus, when middleaged children have a strong preference for their own privacy and avoid coresidence with their frail and disabled elderly parents (Kotlikoff and Morris, 1990), financial assistance to parents might assume an increasingly important role in the transfer of resources from the middle to the oldest generation. References Blank, R. M. (1989). The role of parttime work in women's labor choices over time. American Economic Review, 79, Boaz, R. F., & Miiller, C. F. ( 1992). Paid work and unpaid help by caregivers of the disabled and frail elderly. Medical Care, 30, BoerschSupan, A. H. (1990). Dynamic analysis of household dissolutions and living arrangement transitions by elderly Americans. In D. W. Wise (Ed.), Issues in the economics of aging (pp ). Chicago: The University of Chicago Press. BoerschSupan, A. H., Hajivassiliou,., Kotlikoff, L. J., & Morris J. N. (1992). Health, children, and elderly living arrangements: A multiperiod probit model with unobserved heterogeneity and autocorrelated errors. In D. E. Wise (Ed.), Topics in the Economics of Aging (pp ). Chicago: The University of Chicago Press. Doty, P., Jackson, M. E., & Crown, W. (1998). The impact of female caregivers' employment on the patterns of formal and informal eldercare. The Cerontologist, 38, Ettner, S. L. (1996). The opportunity costs of elder care. The Journal of Human Resources, 31, Greene, W. H. (1995). LIMDEP, version 7.0, user manual. New York: Econometrics Software, Inc. Gujarati, D. N. (1988). Basic econometrics (2nd ed.). New York: McGraw Hill. Harris, L, and Associates. (1986). Problems facing elderly Americans living alone: A national survey. Survey conducted for the Commonwealth Fund Commission on Elderly Living Alone, New York. Himes, C. L. (1994). Parental caregiving by adult women: A demographic perspective. Research on Aging, 16, Juster, F. T., & Stafford, F. P. (1991). The allocation of time: Empirical findings, behavioral models, and problems of measurement. Journal of Economic Literature, 29, Juster, F. T., & Suzman, R. M. (1995). An overview of the Health and Retirement Study. The Journal of Human Resources, Supplement, S7 S56. Killingworth, M. (1983). Labor supply. New York: Cambridge University Press. Kotlikoff, L. J., & Morris, J. N. ( 1990). Why don't the elderly live with their children: A new look. In D. E. Wise (Ed.), Issues in the economics of aging (pp ). Chicago: The University of Chicago Press. Liu, K., Manton, K. G., & Liu, B. M. (1985). Home care expenses for the disabled elderly. Health Care Financing Review, 7(2), McDonald, J. F., & Moffitt, R. A. (1980). The uses of tobit analysis. The Review of Economics and Statistics, 62, McGarry, K., & Schoeni, R. F. (1995). Transfer behavior in the Health and Retirement Study: Measurement and redistribution of resources within the family. The Journal of Human Resources, Supplement, S184 S226. Mincer, J. (1974). Schooling, experience, and earnings. New York: Columbia University Press. Muurinen, J. M. (1986). The economics of informal care. Medical Care, 24, National Institute on Aging. (1993). Living arrangements of the elderly. In Profile of America's Elderly, No. 4. Bethesda, MD: Author. Pezzin, L. E., Kemper, P., & Reschovsky, J. (1996). Does publicly provided home care substitute for family care? Experimental evidence with endogenous living arrangements. The Journal of Human Resources, 31, Ruhm, C. J. ( 1996). Gender differences in employment behavior during late middle age. Journal of Gerontology: Social Sciences, 51B, S11 S17. Schoeni, R. F. (1998). Reassessing the decline in childparent oldage coresidence during the twentieth century. Demography, 35, Sloan, F. A., Picone, G., & Hoerger, T. J. (1997). The supply of children's time to disabled elderly parents. Economic Inquiry, 35, Smith, J. P., & Kingston, R. (1997). Demographic and economic correlates of health in old age. Demography, 34, Social Security Administration. (1998). Income of the population 55 or older, 1996 (SSA Pub. No ). Washington, DC: Author. Soldo, B. J., Wolf, D. A., & Agree, E. M (1990). Family, households, and care arrangements of frail older women: A structural analysis. Journal of Gerontology: Social Sciences, 45B, S238S249. Soldo, B. J., & Hill, M. S. (1993). Intergenerational transfers: Economic, demographic, and social perspectives. In G. L. Maddox and M. P. Lawton (Eds.), Annual Review of Gerontology and Geriatrics, ol. 13, pp New York: Springer. Soldo, B. J. ( 1996). Cross pressures on middleaged adults: A broader view. Journal of Gerontology: Social Sciences, 51B, S271S273. Stone, R. I., & Kemper, P. (1989). Spouses and children of disabled elders: How large a constituency for longterm care reform? The Milbank Quarterly, 67, Stone, R. I., & Short, P. F. (1990). The competing demands of employment and informal caregiving of disabled elders. Medical Care, 28, WhiteMeans, S. I., & Chiang, C. F. (1991). Family choices with managed care for home bound elderly. Southern Economic Journal, 58, Wolf, D. A., & Soldo, B. J. (1994). Married women's allocation of time to employment and care for elderly parents. The Journal of Human Resources, 29, Received September 16, 1998 Accepted September 13, 1999 ol. 39, No. 6, Downloaded from